I was thinking about what I really want from my home automation system, and that is of course for everything that I would do normally to be done for me via automations. Continuing to brainstorm here, walk with me 
I would like for each entity to have an implicit bayesian binary_sensor that indicates, given a specified set of other entities, what the state of the focused entity is likely to be. i.e., given things like the entity’s own history, date, time, day of week, weather, temperature, my location, etc., is my living room light likely to be on or off? Is my thermostat likely to be set to to heating or cooling? This seems like it would be way too much to configure manually, especially for even every light and switch, but if it were built in to HA, it seems plausible.
This could be good for a ‘vacation mode’ where entities can just follow the state that their bayesian sensors think they should be, but it could also be the ultimate set of automations. Given that you’ve recorded enough history for the bayesian sensor to pull data from, nearly every entity could behave automatically as if you were controlling it manually. Your lights would turn on as soon as you thought you wanted them to
I feel like this is an idea that somebody must have had before, but I haven’t seen it here. What do you think?